Skip to content

Dataset · sarulab-speech

Cleaned Mozilla Common Voice 22 at 48 kHz

sarulab-speech/commonvoice22_sidon

Tens of millions of speech audio recordings in 137 languages, denoised and enhanced by the Sidon model.

What's inside

A processed version of the open Mozilla Common Voice 22.0 speech corpus, including 10 million to 100 million fragments across 137 languages. Audio files underwent noise reduction and quality restoration up to 48 kHz using the Sidon model, cut into 21-second chunks, and packaged into WebDataset (.tar.gz) archives under the CC0 1.0 license.

What you can do with it

  • Training high-quality multilingual text-to-speech (TTS) models at 48 kHz
  • Training automatic speech recognition (ASR) systems and audio enhancement models
Tasks
Text → speech
Languages
Abkhazian, Afrikaans, Amharic, Arabic, Assamese, Asturian, Azerbaijani, Bashkir and 4 more
Size
10M–100M rows
Downloads this month
2.7K
Likes
32
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("sarulab-speech/commonvoice22_sidon")